Optimal Experiment Design for Automotive Testing
Feature selection to solve the curse of dimensionality in configuration testing
Spoiled for choice
Extensive testing is conducted by automotive manufacturers to ensure safety and reliability. However, the many customizable features in today's cars make it impractical to test every possible unique configuration. This is a prime example of the so-called curse of dimensionality.
The design of the test system can be optimized effectively if we know which features are important in determining performance for any given test. This can reduce the time spent testing without compromising the integrity of the test process, resulting in a more efficient and eco-friendly operation.
The design of the test system can be optimized effectively if we know which features are important in determining performance for any given test. This can reduce the time spent testing without compromising the integrity of the test process, resulting in a more efficient and eco-friendly operation.
Cutting through the noise
Optimal Feature Selection enabled us to find the most succinct explanations of how the test performance was driven by the features.
Compared to traditional methods for feature selection such as the elastic net and lasso, Optimal Feature Selection found models with both significantly better performance and fewer features.
Optimal Feature Selection achieved peak performance with models roughly 10% the size of those found by the elastic net.
Compared to traditional methods for feature selection such as the elastic net and lasso, Optimal Feature Selection found models with both significantly better performance and fewer features.
Optimal Feature Selection achieved peak performance with models roughly 10% the size of those found by the elastic net.
Optimal Feature Selection reaches peak performance with eight features
Optimizing test design
The knowledge that test performance is driven by a very small number of features motivated creation of a refined and streamlined test process.
The relative importance of features and the incremental cost of testing were used to feed an optimization model, resulting in a new and optimal testing system.
The new system was significantly more lightweight than before, due to the sheer reduction in the number of features identified as relevant in determining test performance.
The relative importance of features and the incremental cost of testing were used to feed an optimization model, resulting in a new and optimal testing system.
The new system was significantly more lightweight than before, due to the sheer reduction in the number of features identified as relevant in determining test performance.
Unique Advantage
Why is the Interpretable AI solution unique?
-
True feature selection
Optimal Feature Selection identifies the smallest set of features needed for optimal performance, unlike traditional feature selection heuristics
-
From predictions to prescriptions
The small set of selected features is the solid foundation that feeds the optimization model determining the best design for the testing system